KPMG withdrew a published report on AI usage after it became clear that the findings contained what appear to be AI-generated hallucinations — meaning the data the report was built on may never have existed. For a Big Four firm staking its advisory reputation on AI expertise, retracting a research document is a significant credibility hit.
The core problem is one builders and researchers should internalize: AI models are particularly unreliable when asked to generate or summarize information about AI itself. The space moves fast, training data is often incomplete or outdated, and models will confidently fill gaps with plausible-sounding statistics that have no real-world basis. Using an LLM to shortcut primary research on AI adoption rates is exactly the kind of task where hallucination risk is highest.
This matters beyond the embarrassment angle. Enterprise reports from firms like KPMG feed into boardroom decisions, investment theses, and policy discussions. If the underlying numbers are fabricated, every downstream decision built on them is compromised. The damage isn't contained to KPMG.
The practical takeaway for anyone producing research or content with AI assistance: never let a model be the source of factual claims — only a processor of verified sources. Any statistic, survey figure, or market data point that an LLM produces should be traced back to a primary source before it goes anywhere near a published document. If you can't verify it, cut it.
Verification workflows aren't optional overhead — they're the minimum viable process for using AI in research contexts responsibly. KPMG's retraction is a public reminder of what skipping that step costs.